Compute Futures Become an Institutional Asset Class

Compute gets cheaper per token and more expensive in aggregate. That paradox is creating a new asset class.

In the second half of 2026, pending regulatory review, two of the world's largest derivatives exchanges will begin trading futures contracts on something that has never had a public price before: a unit of computing power. Intercontinental Exchange, operator of the New York Stock Exchange, announced on July 1 a partnership with NATIVX to launch GPU compute futures based on the COIL Index, a tokenized, energy-normalized benchmark for GPU rental rates. CME Group followed a similar path in May, teaming up with Silicon Data, the GPU market intelligence platform backed by trading firm DRW, to build futures contracts on daily GPU rental-rate indices.

What compute futures actually measure

A futures contract for compute works like one for oil or wheat: a standardized unit, a settlement mechanism, and a public price that both sides agree to. The innovation is not in the derivative structure. It is in the index underlying it. GPU rental pricing has historically been opaque, varying wildly across providers, regions, and contract duration. A one-year H100 lease jumped 38.2% between October 2025 and March 2026 alone.

The COIL Index, developed by NATIVX and licensed from Synova Global, addresses this by tracking tokenized GPU compute prices normalized to a consistent energy unit. It strips out regional power cost disparities. A GPU in Iceland costs less to run than one in California, and it produces four sub-indices covering training, inference, graphics, and connectivity. CME's competing product uses Silicon Data's daily benchmarks, which track actual on-demand rental rates across the global GPU market.

The numbers explain why both CME and ICE are moving simultaneously. Multi-trillion-dollar compute spending with no hedging mechanism is an anomaly in any developed financial market. The price of renting an H100 GPU varies by more than 40% depending on the provider, contract length, and data center location. For a large AI builder running 10,000+ GPUs, that variance translates into hundreds of millions of dollars of unhedged exposure.

The institutional dimension

For institutional investors, the significance goes beyond having a new derivatives product to trade. Compute futures create a price-discovery mechanism for an input that currently lacks one. When a sovereign wealth fund evaluates a data center investment, the single largest variable in the underwriting model, the future price of GPU compute, has had no forward curve. Compute futures change that.

The listing of these contracts alongside its existing power and natural gas futures creates what the exchange calls "a uniquely integrated hedging environment." A data center operator can hedge both GPU costs and electricity costs in the same venue. An energy trader can use compute futures as a proxy for AI-driven electricity demand growth, one of the fastest-growing sources of power consumption on the grid.

This matters for every institutional portfolio that has exposure to AI infrastructure. Pension funds, endowments, and sovereign wealth funds have allocated tens of billions to data center equity and debt over the past 18 months. Without compute futures, those positions carry an unhedged volumetric and pricing risk that no other asset class in the portfolio tolerates.

The paradox and its implications

Token prices have fallen roughly 10× per year for equivalent capability since 2023. A million GPT-4-class output tokens that cost $60 in March 2023 now cost between $0.50 and $3.00 depending on the provider. Inference, not training, now accounts for 55-80% of enterprise GPU spend. The cost per token drops; the total bill grows because usage expands faster than price compresses.

That dynamic, cheaper unit and larger absolute spend, is the fundamental reason compute needs a futures market. Traditional procurement (annual contracts with a cloud provider, take-it-or-leave-it pricing) cannot manage the volatility that comes with 10× annual price declines combined with exponentially growing demand. A futures curve lets an AI company lock in next year's compute costs the same way an airline hedges jet fuel.

Compute is now an asset class, and like every asset class, it needs a public price and a market. By combining our energy-normalized index with ICE's global futures marketplace, we're giving the world's largest new commodity the transparent and regulated venue it has been missing.— Cole Crawford, Founder and Chairman of NATIVX

How the market structure evolves from here

The immediate question is adoption velocity. Commodity futures markets typically take 3-5 years to build sufficient open interest for institutional participation. Compute futures have two advantages: real demand exists today (not speculative), and two competing exchanges will market the product simultaneously. CME's Terry Duffy called compute "the new oil of the 21st century," a framing that if accurate implies a faster adoption curve because the underlying physical market is already enormous and concentrated.

The main obstacle is regulatory. Both exchanges stressed that their products are "pending regulatory review" and "subject to completion of relevant regulatory processes." The CFTC has been active on novel derivatives: in July 2026 it exercised its authority to stay CME's 24/7 crude oil futures. Compute futures face a lighter regulatory burden because the underlying, GPU rental rates, is a straightforward service price index, not a novel asset class like event contracts. But the timeline remains uncertain.

What happens to the compute market a year from now?

Development scenarios